Why I Built an AI Fortune-Telling App With Zero Domain Knowledge
I can't tell you what a Day Pillar is. I don't know the difference between a Heavenly Stem and an Earthly Branch. I've never had my fortune told, never drawn a tarot card, never looked at a palm line and thought anything other than "huh, that's a crease in my hand."
And yet, over 29 days this January and February, I built PanPanMao -- an AI-powered Chinese metaphysics platform with 9 product verticals, 85 API endpoints, 284,000 lines of code, and a working credit economy with real paying users. All of it built solo.
This is the story of why.
The Career That Didn't Scratch the Itch
My career has been a fairly standard big-tech trajectory. Airbnb, where I built fraud detection ML systems. Apple, where I worked on Siri's on-device ML pipeline. AWS, where I worked on the Athena query engine. Then a leap into startup life as CTO at Compute Labs, building multi-agent AI systems for GPU infrastructure financing.
At every stop, I was the "build it" person. Someone else decided what to build. A PM wrote the PRD. A designer handed me Figma files. A growth team figured out conversion funnels. My job was to take a well-defined problem and engineer an excellent solution.
I was good at this. I enjoyed it. But something was missing. I wanted the messy parts. The complete product lifecycle, from "I have an idea" to "someone is paying me money for this," with no one else to hand the hard parts to.
Why Chinese Metaphysics, Specifically
If I built another developer tool or SaaS dashboard, I'd just be doing what I already know. So: Chinese metaphysics. BaZi, astrology, tarot, dream interpretation, palm and face reading. Ancient systems for understanding destiny, personality, and fortune.
Three reasons:
Reason 1: I literally know nothing. This is the point. My hypothesis: can AI completely bridge a domain knowledge gap? Not "can AI help an expert work faster" -- can someone with zero domain expertise build a credible product using AI as their entire knowledge base?
Reason 2: It's not a toy domain. The BaZi system alone has layers upon layers -- Five Elements interactions, Heavenly Stems and Earthly Branches, ten-year luck pillars, Day Master concept. It's a genuine knowledge domain with real depth.
Reason 3: There's real, underserved demand. Hundreds of millions of Chinese-speaking people engage with metaphysics services. The existing digital options are mostly terrible.
The Thesis
If one engineer, augmented by AI tools, can build a real, monetized, multi-product platform in a domain they know nothing about -- in under 30 days -- that says something fundamental about the future of software.
Not "AI can write code." That's old news. AI as the universal domain expert. AI as the thing that lets a single builder operate in any domain.
The Stack
- Monorepo: Turborepo
- Framework: Next.js 16, App Router
- Database and Auth: Supabase
- AI Models: Claude + Gemini -- multi-model
- Payments: Stripe
- Language: TypeScript throughout
- AI Dev Tools: Claude Code and OpenAI Codex as pair programmers
The Approach
- Prototype fast. Build standalone apps for each vertical to test feasibility.
- Consolidate. Merge everything into a unified platform.
- Monetize. Build a credit economy with dried fish treats (because the whole platform is cat-themed).
- Iterate. Watch what users do. Ship improvements daily.
January 21, 2026
On that day, I made my first commit to the unified monorepo. 21 commits by end of Day 1. Over the next 28 days: 1,134 commits. 9 product verticals. A complete credit economy. MediaPipe-powered palm reading. Cron-generated daily content. DFA-based content filtering.
All built by one person who still cannot explain what a Heavenly Stem represents.
This is Part 1 of a three-part series. In Part 2, I break down the technical build. In Part 3, the honest lessons.
Here's where it started: one engineer, zero domain knowledge, and a bet that AI could bridge the gap.